[11] Jacob Andreas - Learning from Language
The Thesis Review
00:00
How to Use Policy Sketches to Speed Up Reinforcement Learning
The policy sketches paper is basically saying look here's a little bit of extra supervision we can collect it's super easy to get. Is this enough to actually substantially change the outlook for a multi task reinforcement learning problem? And you know I think the sort of literature that this is most closely related to is work on options policies from like written Sutton and doing a pre-cup which are basically trying to answer a harder version of that question.
Transcript
Play full episode
